An AI-driven multi-power-source intelligent photovoltaic combiner box and working method

Abstract

The application discloses an AI driving-based multi-power intelligent photovoltaic junction box and a working method, and relates to the technical field of solar photovoltaic accessories.The photovoltaic junction box supports the access of switching power supplies with different voltage grades, and comprises a plurality of independent power access circuits, a photovoltaic access unit, a junction unit, a current shunt, a load cutting-in unit and an AI monitoring module.Through the AI monitoring module combined with reinforcement learning and fuzzy control strategy, dynamic adjustment of current distribution, optimization of load switching, fault prediction and response are realized, the reverse flow problem caused by the inconsistent busbar voltage of the multi-power system is solved, the stability and intelligent level of the system are improved, the construction and operation and maintenance costs are reduced, and the photovoltaic application in the communication machine room and other complex power environments is widely applicable.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of solar photovoltaic accessories, in particular to a multi-power intelligent photovoltaic combiner based on AI driving and a working method. BACKGROUND

[0002] In the communication solar power system technology, the combiner box mainly undertakes the functions of current collection and circuit protection. Generally, it is composed of 1-N photovoltaic inputs, corresponding number of circuit breakers or fuse protection, copper bus, lightning protection unit, power metering unit, communication unit and load switching unit. Generally, one combiner box accesses one switching power supply, which is superimposed into the current switching power supply to provide green energy use with photovoltaic priority.

[0003] However, in the existing communication machine room, due to multi-operator sharing and historical construction, many machine rooms have multiple switching power supply systems for power supply; with the construction of the operator minimal station, there also exists a dual power supply scene of a station site with both switching power supply and new blade power supply.

[0004] For the above multi-power scene, if the existing scheme uses a single combiner box, and the output unit accesses different switching power supplies through multiple wires, due to the fact that the voltages of the two switching power supply busbars cannot be guaranteed to be consistent, it leads to unacceptable scenes such as reverse current or only single power supply working, and cannot be used. If multiple power supplies access the site using multiple sets of photovoltaic equipment, it undoubtedly increases the platform input quantity and maintenance quantity, increases the fault points in the future, and also increases the cost.

[0005] Therefore, how to design an intelligent photovoltaic combiner box that can support independent access of multiple power supplies, avoid the reverse current problem caused by inconsistent busbar voltages, and realize dynamic regulation and control of current distribution and load switching management through intelligent optimization, has become a technical problem to be solved. SUMMARY

[0006] In view of the above problems, the present application provides a multi-power intelligent photovoltaic combiner based on AI driving and a working method, which combines AI-driven fault prediction and control optimization algorithm, and uses advanced technologies such as reinforcement learning and fuzzy control to predict and respond to system faults, thereby providing important technical support for efficient operation of the multi-power photovoltaic system of the communication machine room.

[0007] The present application realizes the above-mentioned purposes through the following technical solutions:

[0008] A multi-power intelligent photovoltaic combiner based on AI driving, which supports access of switching power supplies with different voltage levels, comprising: a plurality of power supply access circuits, and each power supply access circuit works independently and does not interfere with each other, and each independent power supply access circuit comprises:

[0009] A photovoltaic access unit for accessing independent photovoltaic arrays to transmit direct current to a current collection unit;

[0010] A current collection unit for collecting the current of the photovoltaic arrays and uniformly transmitting the current to a current distributor;

[0011] A current distributor for monitoring and distributing the current from the current collection unit;

[0012] A load switching unit for determining the access and switching of the switching power supply according to the load demand and the health status of the loop, supporting any one or all power supply, and automatically switching to the other power supply when any one power supply fails;

[0013] A lightning protection unit for integrating lightning protection to avoid the impact of lightning or voltage surge on the loop;

[0014] The current collection box further comprises an AI monitoring module for detecting the health status of the loop by combining reinforcement learning and fuzzy control strategy, adjusting the current distribution and power switching in real time, adjusting the fuzzy control rules according to real-time data, predicting potential failures, and providing fault diagnosis and response.

[0015] As a preferred scheme of the present application, the current collection box further comprises:

[0016] A grounding unit for providing electrical grounding protection for the current collection box and all power supply access loops;

[0017] A local load power supply unit for providing power support to local loads according to the output power of the loop;

[0018] A multi-loop metering unit for metering the power of each power supply access loop, monitoring the voltage, current and power of each loop in real time, and uploading the data to the communication unit;

[0019] A communication unit for transmitting the data uploaded by the multi-loop metering unit and the operating state information of the current collection box to the AI monitoring module for data analysis, fault diagnosis and optimization.

[0020] As a preferred scheme of the present application, the AI monitoring module comprises:

[0021] A fault detection unit for monitoring the current, voltage and temperature of each power supply access loop in real time, determining whether there is a fault, and feeding back the fault information to the load switching decision unit;

[0022] A load switching decision unit for intelligently determining whether to switch the power supply access loop to ensure stable power supply to the load according to the fault information provided by the fault detection unit and the real-time load demand, and issuing instructions to automatically switch to another power supply through the load switching unit if necessary;

[0023] a current optimization control unit for dynamically adjusting the current distribution according to the real-time current and load demand, and optimizing the current output of each loop through a fuzzy control strategy;

[0024] a learning and prediction unit for analyzing historical operation data and real-time monitoring data through a reinforcement learning algorithm, predicting potential loop faults, and generating an optimized control strategy to adjust the current distribution and load switching decision in advance.

[0025] As a preferred scheme of the present application, the fault detection unit performs fault mode recognition based on a Q-learning algorithm, calculates the Q value of each loop, selects the optimal strategy and dynamically updates, and the specific implementation process is as follows:

[0026] Define the state space S and the action space A, the state space S represents the current, voltage and temperature of each loop, and the action space A represents the fault detection and diagnosis action;

[0027] When the fault detection unit is in the current state s t , select an action a t to determine whether the health state of the loop is normal;

[0028] After performing the action a t , state transition is performed, from the current state s t to the new state s t+1 , and the immediate reward value r t+1 is calculated according to the performed action;

[0029] If the current, voltage or temperature of the loop is within the normal range, the immediate reward value r t+1 is +1, indicating normal operation; if the state exceeds the normal range, a current overload, voltage anomaly or high temperature fault occurs, the immediate reward value r t+1 is -1, indicating a fault state;

[0030] Once the immediate reward value r t+1 is obtained, the Q-learning algorithm update formula is used to update the Q value corresponding to the current state s t and the selected action a t , and the update formula is:

[0031]

[0032] In the formula, Q(s t ,a t ) represents the Q value of performing action a t in state s t ; α is the learning rate; r t+1 is the immediate reward value given according to the health state of the loop; γ is the discount factor; and a' is the next state s t+1The best action in all the following actions;

[0033] Whenever the state changes or the execution of the action produces a different reward, the Q-learning algorithm dynamically adjusts the Q value, so that the fault detection strategy gradually tends to be optimal.

[0034] As a preferred scheme of the present application, the load switching decision unit optimizes the decision based on the multi-armed bandit problem, selects the power switching strategy, and specifically comprises:

[0035] Each power access circuit is regarded as a pull rod, and each selection of any power access circuit for switching brings a reward, and the reward r t represents the stability of the load after power switching, r t = +1 indicates that the load is stable and the power supply meets the demand, r t = -1 indicates that the load is unstable and the power supply is insufficient;

[0036] Based on the current reward and the historical reward, a power access circuit is selected for switching using a greedy strategy or an epsilon-greedy strategy, and the reward value r t is calculated according to the stability of the load power supply after switching.

[0037] By recording the selected power access circuit and the reward each time, the data is gradually accumulated and the reward value is updated, and the reward of each power access circuit is calculated, and the formula is:

[0038]

[0039] In the formula, Q(A i ) represents the reward value of the power access circuit A i ; represents the instantaneous reward after the tth selection when the power access circuit A i is selected; N i is the number of times the power access circuit A i is selected; and T is the total number of decisions.

[0040] According to the updated reward value, the optimal power access circuit is selected, and the power access circuit with the highest reward is selected as the preferred power switching next time. If the rewards of multiple power access circuits are the same, an epsilon-greedy strategy is used for selection.

[0041] The reward of each power switching is recorded, the reward value of each power access circuit is updated, and the selection strategy is adjusted.

[0042] As a preferred scheme of the present application, the current optimization control unit monitors the current, load demand and operation state of each circuit in real time, and evaluates whether the current distribution meets the expectation.

[0043] According to the nonlinear relationship between the current and the load demand, a fuzzy control strategy is adopted to dynamically adjust the current distribution to optimize the current output of each loop;

[0044] The fuzzy control strategy adjusts the current based on fuzzy rules. When the current of any loop is lower than the preset threshold, it is determined whether the current output of the loop needs to be increased according to the fuzzy control rules, and corresponding adjustments are made according to the load demand of other loops.

[0045] As a preferred scheme of the present application, the learning and prediction unit adopts a deep reinforcement learning algorithm to predict potential loop failures and generate an optimized control strategy, including the following steps:

[0046] By analyzing historical failure patterns, current, voltage, temperature, load demand and real-time operation data, a deep reinforcement learning model is established to predict future possible system failures;

[0047] Through deep neural network DNN learning of potential patterns of load switching, failure occurrence and current distribution, Q-learning or Actor-Critic algorithm in the training process, current distribution and load switching decisions are optimized according to historical data to predict the probability of failure occurrence;

[0048] Through the learned strategy and model, the current distribution and load switching decisions are adjusted in advance, the potential risk of failure occurrence is predicted by the deep reinforcement learning model, and the current distribution is adjusted in advance to optimize the load switching decisions.

[0049] A working method of an AI-driven multi-power intelligent photovoltaic combiner box, applied to the AI-driven multi-power intelligent photovoltaic combiner box as described above, the method comprising:

[0050] Each independent photovoltaic array is connected to the combiner box through the photovoltaic access unit, the combiner unit collects direct current energy from each photovoltaic array and transmits it to the current distributor for current monitoring and distribution;

[0051] The current distributor distributes the current according to the current from the combiner unit and monitors the current state of each loop in real time. During the current distribution process, the AI monitoring module adjusts the current distribution in combination with real-time data to ensure that the current output of each loop meets the load demand and avoids overloading or uneven current distribution;

[0052] According to the load demand and the health status of the loop, the load switching unit intelligently judges the access state of the current power supply and automatically switches the power supply according to the prediction or fault diagnosis result. If any power supply fails, it will automatically switch to other available power supply according to the judgment of the AI monitoring module to ensure continuous power supply for the load;

[0053] The AI monitoring module continuously detects the health state of the system loop by combining reinforcement learning and fuzzy control strategy, and adjusts the current distribution and power switching strategy in real time, predicts the occurrence of potential faults by learning historical data and real-time monitoring information, and takes adjustment measures in advance according to the prediction result, reduces the probability of fault occurrence, and ensures the stable operation of the system.

[0054] When a potential fault or load demand change is predicted, the current distribution strategy is dynamically adjusted using fuzzy control rules to optimize the current output of each loop.

[0055] The beneficial effects of the present application are: by connecting multiple independent power sources to the loop, independent operation and non-interference of power sources of different voltage levels are achieved, and the problems of reverse current and single power source operation caused by inconsistent busbar voltage in a multi-power source system are solved; the design of the photovoltaic access unit, the current combiner unit and the current splitter, combined with the intelligent switching function of the load switching unit, can automatically switch power according to load demand and loop health state, ensuring continuous power supply when any power source fails, improving the reliability of the system; the AI monitoring module combines reinforcement learning and fuzzy control strategy to realize dynamic adjustment of current distribution and load switching, and has fault prediction and diagnosis capability, effectively reducing the risk of failure and ensuring stable operation of the system. The lightning protection module and the power metering unit are also integrated, providing circuit protection and power monitoring capability, further reducing operation and maintenance cost through remote monitoring function, and comprehensively improving the intelligence and practicality of the photovoltaic combiner box. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0057] Among them:

[0058] Figure 1 The module structure diagram of the photovoltaic combiner box in the embodiment of the present application;

[0059] Figure 2 The method flow chart in the embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the embodiments of the present application will be described clearly and completely below with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the described embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present application.

[0061] As shown in FIG. 1, an embodiment of the present application provides an AI-driven multi-power intelligent photovoltaic combiner box supporting the access of switching power supplies of different voltage grades, including multiple power access circuits, and each power access circuit works independently without interference. Figure 1

[0062] The embodiment takes dual power as an example, and the input and output shunts are divided into two circuits. The input shunt of each circuit requires flexibility and can be configured. The output shunt is two-way, and the two circuits are independent. Typical configurations are 4+4, 8+8, 4+8, etc. Each independent power access circuit includes:

[0063] A photovoltaic access unit (circuit 1 photovoltaic access unit, circuit 2 photovoltaic access unit) is used to access an independent photovoltaic array and transmit direct current to a current combiner unit.

[0064] A current combiner unit (circuit 1 current combiner unit, circuit 2 current combiner unit) is used to collect the current of the photovoltaic array and uniformly transmit it to a current shunt.

[0065] A current shunt (circuit 1 current shunt, circuit 2 current shunt) is used to monitor and distribute the current from the current combiner unit.

[0066] A load cut-in unit (circuit 1 load cut-in unit, circuit 2 load cut-in unit) is used to determine the access and switching of the switching power supply according to the load demand and the health status of the circuit, support any one-way power or all-way power, and automatically switch to other power when any one-way power fails.

[0067] A lightning protection unit (circuit 1 lightning protection unit, circuit 2 lightning protection unit) is used to integrate lightning protection to avoid the influence of lightning or voltage impact on the circuit. It has an independent SPD module for lightning protection. The nominal operating voltage of the SPD should be greater than 1.2 times the maximum operating voltage of the light control controller. The rated current of the SPD is In≥10kA, and the lightning protection module has a remote signaling function.

[0068] Further including:

[0069] ​An AI monitoring module is configured to detect the health status of the loop by combining reinforcement learning and fuzzy control strategies, to adjust the current distribution and power switching in real time, and to predict potential failures, provide fault diagnosis and response according to real-time data, and adjust the fuzzy control rules according to real-time data.

[0070] The AI monitoring module has the function of accessing the tower monitoring platform, and can monitor the system working state in real time; collect and store system operation parameters; use a wireless transmission module to realize data transmission between the controller and the monitoring platform through the MQTT protocol, and support remote upgrading. The monitoring content is as follows:

[0071] (1) Photovoltaic adapter

[0072] Telemetry: adapter manufacturer, adapter model, adapter software version, adapter on / off state, adapter temperature, adapter input voltage / current, adapter output voltage / current, adapter output power.

[0073] Remote signaling: adapter fault, adapter protection trigger, adapter communication alarm, adapter high temperature alarm.

[0074] (2) Photovoltaic controller

[0075] Telemetry: system type, IMSI, MSISDN, SIM card ICCID, software and hardware version, controller manufacturer, controller model, system total power, cumulative power generation.

[0076] Remote signaling: system output voltage alarm, system lightning protection device state, controller fault alarm.

[0077] Remote control: system zero point time calibration.

[0078] (3) DC ammeter

[0079] Telemetry: meter number, total DC current, DC voltage, total active power, total active energy, daily active energy, daily forward active energy and total forward active energy in each peak and valley period.

[0080] Remote signaling: fault alarm.

[0081] Remote adjustment: start and end time of each peak and valley period.

[0082] (4) Environmental quantity

[0083] Telemetry: temperature, humidity, wind direction, wind speed, irradiance.

[0084] Grounding unit for providing electrical grounding protection for the busbar and all power supply access loops; the protection grounding point should have obvious signs, and the resistance between the shell and all accessible non-live metal parts and the protection grounding point should not be greater than 0.1Ω;

[0085] a local load power supply unit for providing power support to the local load according to the electrical energy output by the circuit;

[0086] a multi-circuit metering unit for metering the electrical quantity of each power access circuit, monitoring the voltage, current and power of each circuit in real time, and uploading to the communication unit; the DC ammeter used by the multi-circuit metering unit shall comply with GB / T 33708-2017 Static DC Energy Meter, has CPA certification, and the accuracy is not less than 1 level, has forward, reverse and time-sharing energy metering functions. The DC ammeter requires:

[0087] a) The ammeter shall have CPA metering certification;

[0088] b) The ammeter shall have forward and reverse energy metering functions;

[0089] c) The ammeter shall be able to measure real-time DC voltage, current and power parameters;

[0090] d) The ammeter shall have an RS-485 communication interface, supporting DL / T645-2007 protocol or Modbus-RTU protocol;

[0091] e) The ammeter auxiliary power supply shall support DC 20V-60V;

[0092] f) The ammeter accuracy is 1 level;

[0093] g) Supports dual-circuit metering capability, can display the voltage, current and energy of two DC branches on the display screen;

[0094] a communication unit for transmitting the data uploaded by the multi-circuit metering unit and the operating state information of the busbar box to the AI monitoring module for data analysis, fault diagnosis and optimization;

[0095] a box body, the box body is firm, the surface is smooth, the color is uniform, there is no rust, wrinkle, flow mark and other defects; the label sign is complete and clear; the lock catch should be easy to operate and flexible and reliable; the warning sign of the box body is marked with the warning sign of the box body. All cables in the busbar box use flame-retardant cables, communication cables and signal acquisition cables shall use shielded cables. The back of the box body is arranged with mounting holes, the specific arrangement position and hole diameter shall be determined in cooperation with the support to ensure smooth installation and fixation.

[0096] Further, the AI monitoring module includes a fault detection unit, a load switching decision unit, a current optimization control unit and a learning and prediction unit.

[0097] The fault detection unit is used to monitor the current, voltage and temperature of each power access circuit in real time, judge whether there is a fault, and feed back the fault information to the load switching decision unit;

[0098] Specifically, the fault detection unit performs fault mode recognition based on a Q-learning algorithm, calculates a Q value of each loop, selects an optimal strategy and dynamically updates, and the specific implementation process is as follows:

[0099] Define a state space S and an action space A, the state space S represents the current, voltage and temperature of each loop, and the action space A represents fault detection and diagnosis actions such as normal operation, current overload alarm, voltage anomaly alarm, etc.

[0100] When the fault detection unit is in a current state s t , an action a t is selected, and it is determined whether the health state of the loop is normal;

[0101] After the action a t is performed, state transition is performed, from the current state s t to a new state s t+1 , and an immediate reward value r t+1 is calculated according to the performed action;

[0102] If the current, voltage or temperature of the loop is within a normal range, the immediate reward value r t+1 is +1, indicating normal operation; if the state exceeds the normal range, current overload, voltage anomaly or high temperature fault occurs, and the immediate reward value r t+1 is -1, indicating a fault state;

[0103] Once the immediate reward value r t+1 is obtained, the Q value corresponding to the current state s t and the selected action a t is updated using the update formula of the Q-learning algorithm, and the update formula is:

[0104] Q(s t ,a t )←Q(s t ,a t )+α[r t+1 +γm a a ′ xQ(s t+1 ,a′)-Q(s t ,a t )];

[0105] In the formula, Q(s t ,a t ) represents the Q value (i.e. the expected return of performing the action) of performing the action a t in the state s t ; α is a learning rate, used to control the influence degree of new experience on Q value update; r t+1For the immediate reward value given according to the health state of the loop, if the loop state is normal, the reward is positive; if a fault (such as overload, overvoltage or overtemperature) is found, the reward is negative; γ is the discount factor, indicating the influence of future rewards on current decision-making; a' is the next state s t+1 The optimal action in all the following actions;

[0106] Whenever the state changes or the execution of the action produces different rewards, the Q-learning algorithm dynamically adjusts the Q value, so that the fault detection strategy gradually tends to be optimal.

[0107] Through the Q-learning algorithm, the system will select an action a t (For example, marked as "normal" or "fault") according to the current loop current, voltage and temperature (i.e. the current state s t Then, the system will observe the state transition (such as current overload, overvoltage, etc.) caused by the execution of the action, and update the Q value according to the new state s t+1 In order to make better fault identification decisions in the future.

[0108] Using the Q-learning algorithm for fault pattern recognition, the optimal detection strategy is selected by dynamically updating the Q value, so that accurate fault judgment and response can be achieved; through the self-learning ability of reinforcement learning, the fault detection unit can continuously optimize the fault detection accuracy according to historical data and real-time monitoring data, improve the intelligent level of the system, efficiently identify faults in the multi-power system, provide accurate feedback and ensure the reliability and stability of the system.

[0109] The load switching decision unit is used to intelligently judge whether the power supply needs to be switched to access the loop to ensure stable power supply to the load according to the fault information provided by the fault detection unit and the real-time load demand, and if so, issue instructions to automatically switch to another power supply through the load switching unit;

[0110] Specifically, the load switching decision unit optimizes the decision-making based on the multi-armed bandit problem and selects the power switching strategy. The multi-armed bandit problem is a classic decision optimization problem, mainly used to deal with the situation of selecting the optimal action (i.e. selecting which "lever") in an uncertain environment. In this problem, it is assumed that there are multiple "levers" (actions), and the reward distribution of each lever is unknown. The goal of the decision maker is to maximize the total reward by selecting these levers multiple times. In the load switching decision, each power loop or backup power supply corresponds to a "lever", and each decision to select a power loop switching will get a "reward", which is directly related to the stability of the load power supply (whether it can meet the load demand) after the power switching.

[0111] Specifically, it includes:

[0112] Each power access circuit is considered as a pull lever, and each time a power access circuit is selected for switching, a reward r t r represents the stability of the load after power switching, r t r = +1 indicates that the load is stable and the power supply meets the demand, r t r = -1 indicates that the load is unstable and the power supply is insufficient;

[0113] Based on the current reward and historical reward, a greedy strategy or ε-greedy strategy is used to select a power access circuit for switching, and the reward value r t is calculated according to the stability of the load after power switching;

[0114] Greedy strategy: select the power circuit with the maximum reward (i.e., stable load and reliable power supply), and each time the action with the maximum Q value is selected;

[0115] ε-greedy strategy: most of the time, select the current optimal power circuit (maximum Q value), occasionally select other circuits for exploration, to ensure that the system constantly tries different power configurations and avoids falling into local optimum;

[0116] By recording the selected power access circuit and reward each time, the data is gradually accumulated and the reward value is updated, and the reward of each power access circuit is calculated, the formula is:

[0117]

[0118] In the formula, Q(A i ) represents the reward value of the power access circuit A i (i.e., the selected "pull lever" or power circuit); r t represents the immediate reward after the tth selection when the power access circuit A i is selected, and the reward value is usually an indication of whether the load power is stable; N i is the number of times the power access circuit A i is selected, i.e., the frequency of the circuit A i being pulled as a "pull lever"; T is the total number of decisions, in the learning process, T represents the time step or the number of selections;

[0119] According to the updated reward value, the optimal power access circuit is selected, and the power access circuit with the highest reward is selected as the preferred power switching next time, if the rewards of multiple power access circuits are the same, the ε-greedy strategy is used for selection;

[0120] The reward of each power switching is recorded, the reward value of each power access circuit is updated, and the selection strategy is adjusted.

[0121] Through the above process, the load switching decision unit can use the optimization method of the multi-armed bandit problem in a multi-power system to select the optimal power switching scheme based on the reward maximization strategy. By balancing exploration and utilization through the ε-greedy strategy, the selection experience of the power loop is continuously accumulated, the power switching decision is optimized, and the stability of the load power supply is ensured.

[0122] The current optimization control unit is used to dynamically adjust the current distribution according to the real-time current and load demand, and optimize the current output of each loop through a fuzzy control strategy.

[0123] Specifically, the current optimization control unit monitors the current, load demand and operating state of each loop in real time, and evaluates whether the current distribution meets the expectations.

[0124] According to the nonlinear relationship between current and load demand, a fuzzy control strategy is used to dynamically adjust the current distribution to optimize the current output of each loop.

[0125] The fuzzy control strategy adjusts the current based on fuzzy rules. When the current of any loop is lower than the preset threshold, it determines whether to increase the current output of the loop according to the fuzzy control rules, and makes corresponding adjustments according to the load demand of other loops.

[0126] The current optimization control unit combines historical data and real-time monitoring data to predict the trend of current demand, and adjusts the current distribution in advance to ensure stable operation of the system and maximize load balancing and system efficiency.

[0127] The learning and prediction unit is used to analyze historical operation data and real-time monitoring data through a reinforcement learning algorithm to predict potential loop failures and generate an optimized control strategy to adjust the current distribution and load switching decision in advance.

[0128] Specifically, the learning and prediction unit uses a deep reinforcement learning algorithm to predict potential loop failures and generate an optimized control strategy, including the following steps:

[0129] By analyzing historical failure patterns, current, voltage, temperature, load demand and real-time operation data, a deep reinforcement learning model is established to predict possible future system failures. The input features of the model include historical failure records and real-time data, which are preprocessed using feature engineering. Deep neural networks (DNN) or other suitable neural network architectures (such as LSTM or CNN) are used for learning to predict possible future system failures.

[0130] Through deep neural network DNN learning of potential patterns of load switching, failure occurrence and current distribution, Q-learning or Actor-Critic algorithm is used in the training process to optimize current distribution and load switching decisions based on historical data, and to predict the probability of failure occurrence.

[0131] Through the learned strategy and model, the current distribution and load switching decision are adjusted in advance, the potential risk of fault occurrence is predicted by the deep reinforcement learning model, the current distribution is adjusted in advance, and the load switching decision is optimized to ensure that preventive control is realized by adjusting the load switching strategy before the fault occurs, thereby reducing the probability of fault occurrence and ensuring stable operation of the system.

[0132] As Figure 2 shown, another embodiment of the present application provides a working method of an AI-driven multi-power intelligent photovoltaic combiner box, applied to the AI-driven multi-power intelligent photovoltaic combiner box as described above, comprising the following steps:

[0133] Photovoltaic array access and current collection: through the photovoltaic access unit, each independent photovoltaic array is accessed to the combiner box, the combiner unit collects the direct current energy from each photovoltaic array, and transmits it to the current diverter for current monitoring and distribution;

[0134] Current distribution and monitoring: the current diverter distributes the current according to the current from the combiner unit and monitors the current state of each loop in real time. During the current distribution process, the AI monitoring module adjusts the current distribution in combination with real-time data to ensure that the current output of each loop meets the load demand and avoids overloading or uneven current distribution;

[0135] Load switching control: according to the load demand and the health status of the loop, the load switching unit intelligently judges the access state of the current power supply, and automatically switches the power supply according to the prediction or fault diagnosis result. If any power supply fails, it will automatically switch to other available power supply according to the judgment of the AI monitoring module to ensure continuous power supply for the load;

[0136] Fault prediction and response: the AI monitoring module continuously detects the health status of the system loop by combining reinforcement learning and fuzzy control strategy, and adjusts the current distribution and power switching strategy in real time. By learning historical data and real-time monitoring information, potential faults are predicted, and adjustment measures are taken in advance according to the prediction results to reduce the probability of fault occurrence and ensure stable operation of the system;

[0137] Current adjustment and optimization control: when potential faults or load demand changes are predicted, the fuzzy control rule is used to dynamically adjust the current distribution strategy to optimize the current output of each loop.

[0138] In summary, the present application solves the problem of reverse current caused by inconsistent busbar voltage in multi-power supply systems by designing multiple power supply access circuits, each of which can work independently and do not interfere with each other, supporting different voltage level switching power supply access. At the same time, it avoids the problem of needing to configure photovoltaic equipment independently for each power supply system in the traditional scheme, reducing the construction and maintenance cost of the system.

[0139] Through the photovoltaic access unit, the current shunt and the current shunt, combined with the AI monitoring module, the current distribution is dynamically adjusted to ensure the balance of the current distribution of each circuit and the efficient output. The load cut-in unit can automatically switch power supply according to the load demand and the health status of the circuit, ensuring continuous and stable power supply of the system when a fault occurs.

[0140] The AI monitoring module detects and predicts the health status of the circuit in real time through reinforcement learning and fuzzy control strategy, dynamically adjusts the current distribution and load switching strategy before the fault occurs, realizes the early prevention and rapid response of the fault, significantly reduces the fault risk in the system operation, and improves the reliability and stability of the photovoltaic system.

[0141] The integrated design of the lightning protection unit effectively protects the system from lightning and voltage impact, providing high-level safety protection for the operation of the photovoltaic system in complex environments.

[0142] The overall technical scheme optimizes the current distribution and power switching strategy by combining historical data and real-time monitoring data, maximizes the power generation efficiency of the photovoltaic system, and flexibly adapts to the dynamic changes of load demand, improving the overall operation efficiency and adaptability of the system, breaking through the limitations of existing technologies, realizing an intelligent, multifunctional and high-reliability photovoltaic combiner box system in a multi-power supply scenario, especially suitable for efficient management of communication machine rooms, minimalist sites and other complex power supply environments.

[0143] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of various changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An AI-driven multi-power intelligent photovoltaic combiner box based on, characterized by, The current collection box supports switch power supply access of different voltage levels, comprising: a plurality of power supply access circuits, and each power supply access circuit works independently and does not interfere with each other, and each independent power supply access circuit comprises: a photovoltaic access unit for accessing an independent photovoltaic array and transmitting direct current to the current collection unit; a current collection unit for collecting the current of the photovoltaic array and uniformly transmitting the current to the current distributor; a current distributor for monitoring and distributing the current from the current collection unit; a load switching unit for determining the access and switching of the switch power supply according to the load demand and the health status of the circuit, supporting any one power supply or all power supplies, and automatically switching to other power supply when any one power supply fails; a lightning protection unit for integrating lightning protection to avoid the influence of lightning or voltage impact on the circuit; The current collection box further comprises an AI monitoring module for detecting the health status of the circuit by combining reinforcement learning and fuzzy control strategy, adjusting the current distribution and power switching in real time, adjusting the fuzzy control rules according to real-time data, predicting potential failures, and providing fault diagnosis and response. The AI monitoring module comprises: a fault detection unit for monitoring the current, voltage and temperature of each power supply access circuit in real time, judging whether there is a fault, and feeding back the fault information to the load switching decision unit; a load switching decision unit for intelligently judging whether the power supply access circuit needs to be switched to ensure stable power supply for the load according to the fault information provided by the fault detection unit and the real-time load demand, and issuing instructions to automatically switch to another power supply through the load switching unit if needed; a current optimization control unit for dynamically adjusting the current distribution according to the real-time current and load demand, and optimizing the current output of each circuit through the fuzzy control strategy; a learning and prediction unit for analyzing historical operation data and real-time monitoring data through a reinforcement learning algorithm, predicting potential circuit failures, and generating an optimized control strategy to adjust the current distribution and load switching decision in advance.

2. The AI-driven multi-power source intelligent PV combiner box of claim 1, wherein The current collection box further comprises: a grounding unit for providing electrical grounding protection for the current collection box and all power supply access circuits; a local load power supply unit for providing power support to local loads according to the output power of the circuit; a multi-circuit metering unit for metering the power of each power supply access circuit, monitoring the voltage, current and power of each circuit in real time, and uploading the data to the communication unit; a communication unit for transmitting the data uploaded by the multi-circuit metering unit and the operating state information of the current collection box to the AI monitoring module for data analysis, fault diagnosis and optimization.

3. The AI-driven multi-power source intelligent PV combiner box of claim 1, wherein The fault detection unit performs fault mode recognition based on the Q-learning algorithm, calculates the Q value of each circuit, selects the optimal strategy and dynamically updates it, and the specific implementation process is as follows: Define the state space S and the action space A, the state space S represents the current, voltage and temperature of each circuit, and the action space A represents the fault detection and diagnosis action; When the fault detection unit is in the current state s t selects the action a t , judges whether the health state of the loop is normal; Perform action a t State transition is performed from the current state s t To a new state s t+1 And the immediate reward value r is calculated according to the performed action t+1 ; If the current, voltage or temperature of the circuit is in the normal range, the instant reward value r t+1 is +1, indicating normal operation; if the state is out of the normal range, a fault of current overload, voltage anomaly or temperature overload occurs, the instant reward value r t+1 is -1, indicating a fault state; Once the immediate reward value r is obtained t+1 The current state s t and the selected action a t are updated using the update formula of the Q-learning algorithm, and the corresponding Q value is updated. The update formula is: Q(s t ,a t )←Q(s t ,a t )+α[r t+1 +γm a a ′ xQ(s t+1 ,a′)-Q(s t ,a t )] where Q(s t , a t ) represents the Q-value of performing action a t in state s t ; a is the learning rate; r t+1 is the immediate reward value given according to the health state of the loop; g is the discount factor; and a' is the best action among all actions in the next state s t+1 . Whenever the state changes or the action produces different rewards, the Q-learning algorithm will dynamically adjust the Q value, so that the fault detection strategy gradually tends to be optimal.

4. The AI-driven multi-power source intelligent PV combiner box of claim 1, wherein The load switching decision unit optimizes the decision based on the multi-armed bandit problem and selects the power switching strategy, which specifically includes: Each power access circuit is considered as a lever, and each time a power access circuit is selected for switching, a reward r t r represents the stability of the load after power switching t r = +1 indicates that the load is stable and the power supply meets the demand t r = -1 indicates that the load is unstable and the power supply is insufficient Based on the current return and the historical return, a power access loop is selected to switch using a greedy strategy or an e-greedy strategy, and a return value r is calculated according to the load power supply stability after switching t ; By recording the power access circuit and the return of each selection, accumulating data step by step and updating the return value, the return of each power access circuit is calculated, and the formula is: where Q(A i ) represents the return value of the power access circuit A i ; r t (i) represents the instantaneous return after the tth selection when the power access circuit A i is selected; N i is the number of times the power access circuit A i is selected; and T is the total number of times of decision making. According to the updated return value, the optimal power access circuit is selected, and the power access circuit with the highest return is selected as the preferred power switching next time. If the returns of multiple power access circuits are the same, the selection is made through an ε-greedy strategy; The return of each power switching is recorded, the return value of each power access circuit is updated, and the selection strategy is adjusted.

5. The AI-driven multi-power source intelligent PV combiner box of claim 1, wherein, The current optimization control unit monitors the current, load demand and operating state of each circuit and power supply in real time, and evaluates whether the current distribution meets the expectations; According to the nonlinear relationship between current and load demand, a fuzzy control strategy is adopted to dynamically adjust the current distribution to optimize the current output of each circuit; The fuzzy control strategy adjusts the current based on fuzzy rules. When the current of any circuit is lower than the preset threshold, it is determined whether the current output of the circuit needs to be increased according to the fuzzy control rules, and the load demand of other circuits is adjusted accordingly.

6. The AI-driven multi-power source intelligent PV combiner box of claim 1, wherein The learning and prediction unit uses a deep reinforcement learning algorithm to predict potential circuit failures and generate an optimized control strategy, including the following steps: By analyzing historical fault patterns, current, voltage, temperature, load demand and real-time operating data, a deep reinforcement learning model is established to predict future potential system failures; Through deep neural network DNN learning of potential patterns of load switching, fault occurrence and current distribution, Q-learning or Actor-Critic algorithm is used in the training process to optimize current distribution and load switching decisions based on historical data to predict the probability of fault occurrence; Through the learned strategy and model, the current distribution and load switching decisions are adjusted in advance, and the potential risk of fault occurrence is predicted by the deep reinforcement learning model, so as to adjust the current distribution in advance and optimize the load switching decisions.

7. A working method of an AI-driven multi-power intelligent photovoltaic combiner box, applied to the AI-driven multi-power intelligent photovoltaic combiner box of any one of claims 1-6, characterized in that, The method comprises: Each independent photovoltaic array is connected to the busbar box through the photovoltaic access unit, and the busbar unit collects direct current from each photovoltaic array and transmits it to the current distributor for current monitoring and distribution; The current distributor distributes the current from the busbar unit and monitors the current state of each circuit in real time. During the current distribution process, the AI monitoring module adjusts the current distribution based on real-time data to ensure that the current output of each circuit meets the load demand and avoids overloading or uneven current distribution; According to the load demand and the health status of the circuit, the load switching unit intelligently judges the access state of the current power supply and automatically switches the power supply according to the prediction or fault diagnosis result. If any power supply fails, it will automatically switch to other available power supply according to the judgment of the AI monitoring module to ensure continuous power supply for the load; The AI monitoring module continuously detects the health status of the system circuit by combining reinforcement learning and fuzzy control strategy, and adjusts the current distribution and power switching strategy in real time. By learning historical data and real-time monitoring information, potential faults are predicted, and adjustment measures are taken in advance according to the prediction results to reduce the probability of fault occurrence and ensure stable operation of the system; The current distribution strategy is dynamically adjusted by using fuzzy control rules when potential faults or load demand changes are predicted, optimizing the current output of each loop.

Images